PIE not handled properly in pgconvert about perfgrind HOT 12 CLOSED

This should be fixable :)

from perfgrind.

Did you profile contains any hits to the executable binary itself? I've got only hits to libc.so and ld-linux.so, but after increasing frequency and disabling software events managed to get hit to the executable itself.

I'm afraid the workload we're profiling is just too small to collect any meaningful data.

from perfgrind.

Did you profile contains any hits to the executable binary itself?

Yes, but not when I retried that.

I've got only hits to libc.so and ld-linux.so, but after increasing frequency and disabling software events managed to get hit to the executable itself.

Seen that on re-testing. But even with a higher frequency and no software event and binaries in the profile I'm back to the original problem: objdump is called with address that doesn't exist because pgconvert put that into the cachegrind profile. Do you see the same?

from perfgrind.

Nope, everything looks to be working fine for me. Can you provide me with your binary, pgdata and grind files?

from perfgrind.

Sure. check.zip

from perfgrind.

Well, looks to be working fine for me:

from perfgrind.

Does that include the machine code for all functions in pginfo_dbg? This is where the error is we track under this issue is most visible.

from perfgrind.

Ok, got it finally. Your executable is PIE, this is not handled properly in pgconvert.

from perfgrind.

Is this issue likely to stay open or can you have a look at this in the next weeks? In the former case we should update the README, possibly linking this issue to document this regression.

from perfgrind.

Ok, I was wrong: non-PIE executables have wrong instructions dump as well.

from perfgrind.

Should be fixed by 0a8fd1c, please test.

from perfgrind.

I didn't checked that specific commit but do so that current master fixed that.

from perfgrind.